Most AI citations are earned, not owned.

The comfortable version of GEO says: structure your own website well, answer questions clearly, add schema, and the assistants will cite you. All of that is necessary — and this site practises it. But the citation data says it is not sufficient. In Muck Rack’s What Is AI Reading? study — more than a million links cited across the major assistants in the second half of 2025 — 82% of citations went to earned media: news coverage, independent reviews, industry publications, third-party content. Owned websites earned a minority of the trust.

Why models prefer third parties

For the same reason a careful buyer does. A brand saying ‘we are excellent’ is a claim; a journalist, a reviewer, or an industry body saying it is evidence. Assistants are built to minimise the risk of repeating self-praise, so they weight independent sources — the same pattern we documented in the dead internet analysis, where the assistants’ citations concentrated on verifiable human sources like Wikipedia and established media. The machine, in other words, reads like a sceptic.

GEO is part PR — and always was

The practical consequence: the discipline of being recommended by AI is roughly half on-site work and half reputation work. On-site: answer pages, structured data, entity clarity, a site machines can parse — the foundation that makes citation possible. Off-site: coverage in publications the models already trust, reviews, expert commentary, data other people want to cite. A brand with a perfect website and no third-party footprint is a well-organised claim; a brand present in independent sources is a corroborated fact.

What this looks like in practice

Publish things worth citing — original data, honest analyses, a named methodology — and place them where journalists and niche publications look. One well-placed industry article can outweigh ten optimised landing pages in the citation economy. This is why we treat media outreach and citable research as part of GEO delivery, not a separate PR line item; and it is why the FIB AI Visibility Check looks at who else talks about you, not just what your website says.

How FIB approaches it

We build both halves: the machine-readable site and the third-party footprint that makes assistants confident enough to name you. If AI answers cite your competitors’ coverage while your excellent website sits unread, get in touch.

Sources

Muck Rack, What Is AI Reading?, updated December 2025 (1M+ cited links across ChatGPT, Claude, Gemini, Perplexity) · Graphite, 2025 · Pew Research Center, 2025.